dkoterwa/kor-sts-cosine-embedding-loss

---

license: cc-by-sa-4.0 dataset_info: features: - name: id dtype: int64 - name: genre dtype: string - name: sentence1 dtype: string - name: sentence2 dtype: string - name: score dtype: float64 splits: - name: train num_bytes: 1034815 num_examples: 5691 - name: valid num_bytes: 297254 num_examples: 1465 - name: test num_bytes: 247409 num_examples: 1376 download_size: 837346 dataset_size: 1579478

Korean Semantic Textual Similarity (KorSTS) Dataset (with converted scores)

For a better dataset description, please visit this GitHub repository prepared by the authors of the article: LINK

This dataset was prepared by converting this dataset Specifically, I've changed the values of scores. If the score was equal or above 2.5, then it was assigned with label "1" (similar). In other cases, label "-1" was assigned (not similar).

I've made these changes in order to apply CosineEmbeddingLoss from PyTorch library LINK.

How to download

from datasets import load_dataset
data = load_dataset("dkoterwa/kor-sts-cosine-embedding-loss")

If you use this dataset for research, please cite this paper:

@article{ham2020kornli,
  title={KorNLI and KorSTS: New Benchmark Datasets for Korean Natural Language Understanding},
  author={Ham, Jiyeon and Choe, Yo Joong and Park, Kyubyong and Choi, Ilji and Soh, Hyungjoon},
  journal={arXiv preprint arXiv:2004.03289},
  year={2020}
}